/plushcap/analysis/langchain/langchain-custom-agents

Custom Agents

What's this blog post about?

The text discusses the introduction of new abstractions for creating custom agents in Python and TypeScript modules. A BaseSingleActionAgent is presented as the highest level abstraction for an agent that can be used in their current AgentExecutor, with a more practical LLMSingleActionAgent also introduced. This LLMSingleActionAgent is highly modular and customizable, consisting of four parts: PromptTemplate, LLM, stop sequence, and OutputParser. The text also outlines future directions for the development of agents, including SingleActionAgents using embeddings or a ConstitutionalChain, as well as Multi-action agents and Plan-execute agents.

Company
LangChain

Date published
April 3, 2023

Author(s)
-

Word count
742

Hacker News points
None found.

Language
English


By Matt Makai. 2021-2024.